Annotated data processing method and device

Abstract

The invention discloses an annotated data processing method, an annotated data processing device, computer equipment and a computer storage medium, relates to the technical field of data annotation, and aims to adjust the distribution of to-be-annotated data and improve the effect of annotated data on model training. The method comprises the steps of obtaining sample data under each category randomly extracted from machine data to serve as to-be-labeled data; in a process of labeling the to-be-labeled data based on a labeling platform, counting labeled data under each category, and judging whether the labeled data under each category respectively reaches a training standard preset for a classification prediction model or not; if yes, taking the labeled data reaching the training standard category as training data, and inputting the training data into the network model for training to obtain a classification prediction model; and updating the to-be-labeled data according to the prediction probability of the classification prediction model for the test data.

Description

technical field [0001] The present invention relates to the technical field of data labeling, in particular to a processing method, device, computer equipment and computer storage medium for labeling data. Background technique [0002] In recent years, with the continuous development of computer and Internet technology, various intelligent applications have emerged one after another, and tools such as big data and artificial intelligence have been gradually applied to practice. Natural language processing is a direction of artificial intelligence, enabling computers to understand human language and understand the content, thoughts and emotions expressed in the language. [0003] Since the mainstream technology of natural language technology processing is mainly based on statistical machine learning, these technologies mainly rely on two aspects, one is the statistical model and optimization algorithm for different tasks; the other is the corresponding large-scale corpus. Th...

AI Technical Summary

Problems solved by technology

In many tasks, the annotation system of the corpus is often difficult to grasp. If the categories are too coarse, the language cannot be described comprehensively and meticulously. If the categories are too fine, the annotation information will be too large and the annotation efficiency will be reduced. In both labeling and named entity recognition tasks, a large amount of labeled data is required to be combined as training data

[0004] Existing labeling systems and labeling methods support multi-person real-time labeling systems, all of which import labeling data at one time and hand it over to labelers. Since the distribution of labeling data is not known before labeling, the extracted data to be labeled is often unbalanced As a result, some important data that needs to be labeled have not been extracted, and some data has been extracted too much and repeated labeling, making the distribution of the data to be labeled uneven, affecting the effect of the labeled data on model training

Images